Fugu-Ultra v1.1
#35 in Open-Source LLMsSakana Ai · v1.1 · ultra · 3× · last seen Jul 25, 2026
32
Momentum
Fugu-Ultra v1.1 is an update to its orchestration model Fugu Ultra released by Sakana AI on July 24, 2026. It is not a single trained language model, but rather a multi-agent orchestration system that dynamically coordinates multiple frontier LLMs (e.g., Claude, GPT, Gemini) and provides them through a single OpenAI-compatible API. Sakana reports improvements of up to 7.9 Benchmark points over v1.0, particularly in ProgramBench and Terminal Bench 2.1, as well as positioning above Anthropic's Fable 5 in complex Coding and Reasoning tasks. The service is proprietary and is offered at the unchanged price of v1.0; according to Sakana, the EU/EEA is currently not served.
Momentum trend
01.05.30.07.
Features
| Key Benchmark (%) | Up to +7.9 points over v1.0, strongest gains on ProgramBench and Terminal Bench 2.1; claimed to beat Fable 5 |
| Context Window (Tokens) | 1,000,000 tokens (standard rate up to 272,000 tokens, higher pricing above) |
| License | Proprietary/hosted (not open source); GitHub repo contains only setup materials and technical report |
| Multimodality | Text and image input (vision) |
| Platform | OpenAI-compatible API (api.sakana.ai/v1), console.sakana.ai, plus Claude Code-compatible interface |
| Price per 1M Tokens | $5 input / $30 output / $0.50 cached input (up to 272K context); above that $10/$45/$1.00 – unchanged from v1.0 |
| Release Date | July 24, 2026 |